DeepSkin 🩹

📝 About the Project

Skin cancer is one of the most common forms of cancer, and early detection is crucial for effective treatment. However, many people worldwide lack easy access to dermatologists, leading to delays in diagnosis and unnecessary medical visits.

DeepSkin aims to bridge this gap by providing an AI-powered solution for analyzing skin lesions. Users can upload images of their skin lesions, and our deep learning model will assess whether the lesion is likely to be malignant or benign.

By filtering out benign cases, DeepSkin helps reduce unnecessary dermatology visits while ensuring that high-risk cases receive urgent medical attention. This not only improves patient outcomes but also optimizes healthcare resources by allowing medical professionals to focus on cases that require immediate intervention.

🩺 Our Team

The DeepSkin Team is composed of 4 members:

• Hansen Julien
• Vermeylen Clément
• Arsanov Ramzan
• Seyfullah Ural

🎯 Objectives

1. Develop a deep learning model to classify skin lesions with high accuracy.
2. Provide an intuitive and user-friendly web application for users to interact with the model.
3. Ensure the system can scale and be deployed efficiently on the cloud.
4. Continuously improve the model by incorporating user feedback and new data.

🛠️ Solution

Core Features

• Image Upload: Users can upload a photo of their skin lesion.
• AI Diagnosis: The deep learning model analyzes the image and classifies it as malignant or benign.
• Confidence Score: The model provides a probability score for its prediction.
• User Feedback Mechanism: Users can report incorrect predictions, improving the model over time.

📊 Data & Feasibility

Data

We are using the HAM10000 dataset from Kaggle: HAM10000 Dataset

Team Expertise

• All team members have a solid understanding of deep learning concepts.
• Experience with CNN architectures and computer vision tasks.

Infrastructure

• Cloud Deployment: The model will be hosted on Google Cloud to ensure accessibility and scalability.
• Backend:
• Frontend:

📈 Metrics

To assess our model's performance, we will evaluate:

• Accuracy: Percentage of correctly classified lesions.
• Precision & Recall: To balance false positives and false negatives.
• ROC Curve & AUC Score: To measure how well the model distinguishes between malignant and benign cases.

🤔 Inference

Inference will be performed in real-time using the deployed model on Google Cloud. Users will receive predictions within seconds of uploading an image.

🔎 Evaluation & Continuous Learning

• User Reports: Collect user feedback on incorrect predictions.
• Continuous Learning: Periodically update the model with new labeled data to improve performance.

🧱 Building Blocks

ID	Week	Task Description	Status	Location	Required/Optional
1.1	W01	Form a team.	✅	Our Team	Required
1.2	W02	Select use case.	✅	USECASE.md	Required
1.3	W02	Define use case.	✅	USECASE.md	Required
1.4	W02	Pick a creative project name.	✅	DeepSkin	Required
1.5	W02	Set up a communication channel.	✅	Discord	Required
1.6	W02	Create a GitHub repository for code versioning.	✅	DeepSkin	Required
1.7	W02	Submit the project card with basic details for feedback.	✅	-	Required
2.1	W03	Perform Exploratory Data Analysis (EDA).	✅	Here	Required
2.2	W03	Set up Cloud environment (create project, grant access, set up billing).	✅	-	Required
2.3	W04	Train your ML model.	✅	Training	Required
2.4	W04	Evaluate your ML model.	✅	Prediction	Required
2.5	W03-W04	Document data analysis and model performance.	✅	-	Required
3.1	W05	Build an API to serve your ML model. Run it locally.	✅	-	Required
3.2	W05	Package the API in a Docker container. Run it locally.	✅	-	Required
3.3	W06	Deploy the API in the Cloud, allowing remote predictions.	✅	-	Required
4.1	W08	Build an automated pipeline for training & deployment (e.g., Kubeflow, Sagemaker, GCP Vertex).	✅	-	Optional
5.1	W09	Run model training as a Cloud job (e.g., on a VM or managed service).	❌	-	Optional
5.2	W10	Build and deploy a simple UI/dashboard to showcase results.	✅	-	Optional
6.1	W10	Build a CI/CD pipeline (e.g., GitHub Actions) with at least one automated step.	✅	-	Required
6.2	W10	CI/CD step: Auto-deploy model serving components.	✅	-	Optional
6.3	W10	CI/CD step: Run Pylint for code quality checks.	✅	-	Optional
6.4	W10	CI/CD step: Run Pytest for unit tests.	⏳	-	Optional